A neural network clamping force model for bolt tightening of wind turbine hubs

Industrial manufacturing of large-scale wind turbines requires the accurate tightening of multiple bolts and nuts, which connect the ball bearings - supporting wind turbine blades - with the hub, a huge mechanical component supporting blades pitch motion. An accurate tightening of bolts and nuts requires uniformly distributed clamping forces along flanges and surfaces of contact between hub and bearings. Due to the role of friction forces and the dynamics of the phenomenon, this process is nonlinear and currently performed manually; it is also time consuming, requiring high-cost equipment and expert operators. This paper proposes a set of neural networks, which infer the clamping force achievable with a tightening tool while fastening M24 nuts on bolts. The tool embeds a torque sensor and shaft encoder, therefore two types of inputs of the neural networks are considered in order to fit the clamping force output: the time signals of (a) the applied torque of the tool and (b) the combination of the torque and of the angular speed of the tool. According to results, neural networks properly model the clamping force, both during the training stage and when exposed to unseen testing data. This approach could be generalized to other industrial processes and specifically to those requiring repetitive tightening tasks and involving highly nonlinear aspects, such as friction forces.